How to rolling restart pods without changing deployment yaml in kubernetes?

Question

In kubernetes there is a rolling update (automatically without downtime) but there is not a rolling restart, at least i could not find. We have to change deployment yaml. Is there a way to make rolling "restart", preferably without changing deployment yaml?

Asa

Answer 1

Before kubernetes 1.15 the answer is no. But there is a workaround of patching deployment spec with a dummy annotation:

kubectl patch deployment web -p \
  "{\"spec\":{\"template\":{\"metadata\":{\"annotations\":{\"date\":\"`date +'%s'`\"}}}}}"

As of kubernetes 1.15 you can use:

kubectl rollout restart deployment your_deployment_name

CLI Improvements

• Created a new kubectl rollout restart command that does a rolling restart of a deployment.
• kubectl rollout restart now works for DaemonSets and StatefulSets

Sulfatize

(kubectl rollout restart works by changing an annotation on the deployment's pod spec, so it doesn't have any cluster-side dependencies; you can use it against older Kubernetes clusters just fine.) – Licht

So having locally installed kubectl 1.15 you can use this on a 1.14 cluster? – Olimpia

@NielsBasjes Yes you can use kubectl 1.15 with apiserver 1.14. Here is more detail on kubernetes version skew policy: kubernetes.io/docs/setup/release/version-skew-policy – Sulfatize

If I do the rolling Update, the running Pods are terminated if the new pods are running. But my pods need to load configs and this can take a few seconds. In these seconds my server is not reachable. Can I set a timeout, when the running pods are termianted? – Punkie

@B.Stucke you can use "terminationGracePeriodSeconds" for draining purpose before termination. But I think your prior need is to set "readinessProbe" to check if configs are loaded. – Sulfatize

I don't have deployments, I have svc/pods - Is there a way to restart pods? – Wilhelm

@Wilhelm every service is backed by a deployment, kubectl get deployment -l app=<service-name> will find the deployment name, then you can rollout-restart that – Alius

Here is the PR that implements rollout restart: github.com/kubernetes/kubernetes/pull/76062 – Annulus

Answer 2

If you use k9s, the restart command can be found if you select deployments, statefulsets or daemonsets:

Gabble

yes, however, I tried this twice this week, and found that it immediately kills the pods, and then starts new ones, thus potentially causing an outage. That doesn't seem like a "rolling restart" as I've seen them. That said, I'm not sure why k9s does that, given this code in the PR that implements it: github.com/derailed/k9s/pull/345/… – Annulus

I think that is because you have your deployment configured to allow that. There are settings that controls how many pods may be killed at once in a rolling upgrade: kubernetes.io/docs/concepts/workloads/controllers/deployment . – Gabble